core/command_queue_mt: Fix crash/hang when buffer fills up

This patch fixes two related issues. One is the race condition in issue #42107..
The other is a crash which happens when the reader is lapped near the end of the buffer.

Backport of 48e8da4 to 3.2

core/command_queue_mt.cpp

@@ -80,7 +80,7 @@ CommandQueueMT::SyncSemaphore *CommandQueueMT::_alloc_sync_sem() {
 
 bool CommandQueueMT::dealloc_one() {
 tryagain:
-	if (dealloc_ptr == write_ptr) {
+	if (dealloc_ptr == (write_ptr_and_epoch >> 1)) {
 		// The queue is empty
 		return false;
 	}
@@ -104,8 +104,8 @@ tryagain:
 
 CommandQueueMT::CommandQueueMT(bool p_sync) {
 
-	read_ptr = 0;
-	write_ptr = 0;
+	read_ptr_and_epoch = 0;
+	write_ptr_and_epoch = 0;
 	dealloc_ptr = 0;
 	mutex = Mutex::create();
 

core/command_queue_mt.h

@@ -316,8 +316,8 @@ class CommandQueueMT {
 	};
 
 	uint8_t *command_mem;
-	uint32_t read_ptr;
-	uint32_t write_ptr;
+	uint32_t read_ptr_and_epoch;
+	uint32_t write_ptr_and_epoch;
 	uint32_t dealloc_ptr;
 	uint32_t command_mem_size;
 	SyncSemaphore sync_sems[SYNC_SEMAPHORES];
@@ -330,7 +330,11 @@ class CommandQueueMT {
 		// alloc size is size+T+safeguard
 		uint32_t alloc_size = ((sizeof(T) + 8 - 1) & ~(8 - 1)) + 8;
 
+		// Assert that the buffer is big enough to hold at least two messages.
+		ERR_FAIL_COND_V(alloc_size * 2 + sizeof(uint32_t) > command_mem_size, NULL);
+
 	tryagain:
+		uint32_t write_ptr = write_ptr_and_epoch >> 1;
 
 		if (write_ptr < dealloc_ptr) {
 			// behind dealloc_ptr, check that there is room
@@ -362,8 +366,13 @@ class CommandQueueMT {
 				// zero means, wrap to beginning
 
 				uint32_t *p = (uint32_t *)&command_mem[write_ptr];
-				*p = 0;
-				write_ptr = 0;
+				*p = 1;
+				write_ptr_and_epoch = 0 | (1 & ~write_ptr_and_epoch); // Invert epoch.
+				// See if we can get the thread to run and clear up some more space while we wait.
+				// This is required if alloc_size * 2 + 4 > COMMAND_MEM_SIZE
+				if (sync) {
+					sync->post();
+				}
 				goto tryagain;
 			}
 		}
@@ -377,6 +386,7 @@ class CommandQueueMT {
 		// allocate the command
 		T *cmd = memnew_placement(&command_mem[write_ptr], T);
 		write_ptr += size;
+		write_ptr_and_epoch = (write_ptr << 1) | (write_ptr_and_epoch & 1);
 		return cmd;
 	}
 
@@ -402,17 +412,19 @@ class CommandQueueMT {
 	tryagain:
 
 		// tried to read an empty queue
-		if (read_ptr == write_ptr) {
+		if (read_ptr_and_epoch == write_ptr_and_epoch) {
 			if (p_lock) unlock();
 			return false;
 		}
 
+		uint32_t read_ptr = read_ptr_and_epoch >> 1;
 		uint32_t size_ptr = read_ptr;
 		uint32_t size = *(uint32_t *)&command_mem[read_ptr] >> 1;
 
 		if (size == 0) {
+			*(uint32_t *)&command_mem[read_ptr] = 0; // clear in-use bit.
 			//end of ringbuffer, wrap
-			read_ptr = 0;
+			read_ptr_and_epoch = 0 | (1 & ~read_ptr_and_epoch); // Invert epoch.
 			goto tryagain;
 		}
 
@@ -422,6 +434,8 @@ class CommandQueueMT {
 
 		read_ptr += size;
 
+		read_ptr_and_epoch = (read_ptr << 1) | (read_ptr_and_epoch & 1);
+
 		if (p_lock) unlock();
 		cmd->call();
 		if (p_lock) lock();